Gatez vs Portkey
Portkey is a best-in-class AI gateway for multi-model routing and prompt engineering.
Gatez is a unified, fully self-hosted L1+L2+L3 platform for regulated teams.
Portkey
A best-in-class AI gateway for multi-model LLM routing — 1,600+ models across 40+ providers with observability, guardrails, caching, and a mature prompt-management studio. Now part of Palo Alto Networks (Prisma AIRS) as of 2026.
SaaS-first, with an open-source gateway proxy and an enterprise hybrid option. Governs LLM traffic — not your general API estate — and adds MCP plus basic agent proxying.
Gatez
A three-layer unified gateway (L1 API + L2 AI + L3 Agent) with multi-tenant governance, built for platform teams who must manage APIs, LLM calls, and agents in one self-hosted control plane.
Combines APISIX, a custom Rust AI Gateway, and a custom Rust Agent Gateway with tenant isolation, per-hop agent identity, and a customer-owned audit trail — fully on-prem, air-gap capable.
Feature Comparison
| Feature | Portkey | Gatez |
|---|---|---|
| Gateway Layers | ||
| API Gateway (L1) — general API traffic | ✗ LLM-only | ✓ APISIX |
| AI Gateway (L2) | ✓ | ✓ Rust |
| Agent Gateway (L3) | ~ basic | ✓ Rust |
| AI Features | ||
| Multi-model routing | ✓ 1,600+ models | ✓ 13 providers |
| Semantic caching | ~ Pro / Enterprise | ✓ Redis + Qdrant, on-prem |
| PII redaction (pre-LLM) | ✓ partner-based | ✓ Presidio, local |
| Per-tenant token budgets | ~ Enterprise-only | ✓ per-tenant |
| Prompt management / versioning | ✓ Prompt Studio | ✗ |
| Local / on-prem LLM (Ollama) | ✓ custom host | ✓ |
| Multi-tenancy & Governance | ||
| Infrastructure-level tenant isolation | ~ workspace RBAC | ✓ tenant_id through every layer |
| Per-tenant rate limiting | ~ virtual-key limits | ✓ Redis sliding window |
| HITL approval gates | ✗ blog concept | ✓ first-class queue |
| Tool allowlists | ~ tool RBAC | ✓ deny-by-default |
| Agent Features | ||
| MCP Protocol | ✓ | ✓ |
| A2A Protocol | ~ proxy + RBAC | ✓ policy-governed |
| Per-hop agent identity (RFC 8693) | ✗ | ✓ token exchange, delegation chain |
| A2A delegation policies + loop detection | ✗ | ✓ cross-tenant block, depth limit |
| A2A topology visualization | ✗ | ✓ |
| Observability & Compliance | ||
| Customer-owned audit store | ✗ SaaS-hosted logs | ✓ ClickHouse on-prem |
| Log retention control | ~ 3–30d; custom = Enterprise | ✓ you set the TTL |
| Cross-layer tracing (L1→L2→L3) | ✗ no L1 | ✓ one OTel trace |
| Usage analytics | ✓ SaaS dashboards | ✓ ClickHouse dashboards |
| Operator + Developer portals | ~ SaaS dashboard | ✓ two self-hosted portals |
| Deployment & Governance | ||
| Fully on-prem (control plane included) | ~ hybrid: SaaS control plane | ✓ |
| Air-gapped deployment | ~ Enterprise custom | ✓ out of the box |
| No external phone-home | ✗ hybrid re-syncs to SaaS | ✓ |
| Ownership | Palo Alto Networks | Independent |
✓ full support · ~ partial / tier-gated · ✗ not available. Portkey capabilities as of mid-2026; verify current tiers on portkey.ai.
Where Portkey Leads
-
LLM provider breadth: 1,600+ models across 40+ providers through one API — materially ahead of Gatez's 13-provider L2
-
Prompt Engineering Studio: versioning, publish/rollback, multi-model A/B testing and a playground — a workflow Gatez does not have
-
SaaS onboarding speed: three lines of code, free tier, no infrastructure — a developer can be in production in minutes
-
Production pedigree: 1T+ tokens/day across 24,000+ organizations — a deep reference base for enterprise evaluations
-
Guardrail & framework ecosystem: 50+ partner guardrails and native SDK hooks for LangGraph, CrewAI, and the OpenAI Agents SDK
-
Palo Alto Networks backing: for teams already on Prisma / Cortex, integration into the PANW security stack can be a procurement shortcut
Where Gatez Leads
-
Unified L1+L2+L3: one control plane for your general API estate (APISIX), LLM calls, and agents — Portkey governs LLM traffic only and cannot replace your API gateway
-
True full on-prem: the entire stack — portals, ClickHouse, Redis, OTel — runs in your network with no phone-home, even in air-gapped environments
-
Customer-owned audit store: ClickHouse runs in your environment — you own the data and set retention, with no dependency on a vendor's infrastructure
-
Per-hop agent identity (RFC 8693): every delegation hop carries a cryptographically verifiable chain — who, acting as whom, for whom — enforced at the gateway. No competitor ships this today
-
Gateway-enforced HITL: tool execution is blocked at L3 and routed to a first-class approval queue in both portals — not delegated to the application
-
Infrastructure-level multi-tenancy: tenant_id flows through APISIX rate-limit buckets, Redis keyspaces, and ClickHouse row filters — provable isolation, not just RBAC
-
Vendor independence: Gatez is not owned by a $100B security vendor — no PANW roadmap or procurement dependency in your AI stack
When to Choose Which
Choose Portkey when:
- → Your primary need is multi-LLM routing across the broadest possible model catalog, without strict data-residency requirements
- → Your team iterates heavily on prompts and needs production-grade prompt versioning and A/B testing
- → You want SaaS onboarding speed with no infrastructure to run
- → You're already a Palo Alto Networks / Prisma customer and want it in your existing agreement
- → Your agent use case is LLM observability and cost governance — not gateway-enforced HITL or per-hop identity
- → You're a startup or mid-market team without an on-prem or air-gap requirement
Choose Gatez when:
- → You're in a regulated environment (banking, healthcare, government, defense) where all data must stay in your infrastructure
- → You need one control plane for your API estate + AI + agents — Portkey can't replace your API gateway
- → You require provable infrastructure-level tenant isolation, not application-layer access control
- → You're deploying air-gapped, where even a SaaS control plane is unacceptable
- → You need gateway-enforced HITL and per-hop RFC 8693 agent identity for auditable multi-agent workflows
- → Your audit trail must live in a database you own, with retention you control